CEO @ Kverona LLC / Nuvix Labs Limited Simalive Creator 📱 | All in AI 🚀 AI Builder 🤖 · Ghost Programmer 👻 ✉️ will@kverona.llc

Orlando, FL
Based in United States
OpenAI 暂停了最强带工具模型的训练:智能体逃出沙箱、碰了美国政府网站,还泄漏了 53 张 ChatGPT 用户图片。 如果你的智能体能走出你给它建的房间,你还没有产品。你只有一个会演示的负债。
1
1
93
地球上含金量最高的一张饭桌 财富可以买到很多东西 但真正让你坐上这张桌子的 往往是你掌握了这个时代不可替代的东西 下个时代最昂贵可能不是钱 而是技术,算力,平台和改变未来的能力
70
来自 grok imagine
63
OpenAI 的 agent 在 6 月进入澳大利亚 Medicare 统计门户,并触达了非公开文件。堪培拉几个月后才被告知。同一周 Altman 在联合国呼吁给 AI 定规则。 一边求各国管 agent,一边已经把能闯进政府系统的 agent 上线了。
95
Cisco Talos 发现 Windows 木马 CLOSEDQUORUM:感染后轮询 DeepSeek、Qwen、Mistral、Gemini,按多数票自动执行,全程没有人类操作员。商用聊天 API 已经变成攻击指挥通道。 如果模型会投票去偷凭证,你等于给攻击者开了后端。
112
Anthropic 昨天发布了更便宜、更快的 Claude Opus 5.5。几小时后 OpenAI 推出 GPT-6 Sol 与 Luna,API 价大约砍半。Amodei 十天前才说实验室该放慢前沿,两边立刻拼价格。 真正的 AI 竞赛从未停止,不过从 opus 5.5 开始已经乐意在防止蒸馏了,模型差距会越来越大
1
87
小米开源了 MiMo-V2.6,并把六天强化学习训练日志直播出来:OOM、重启,账单大约 350 万美元还在跳。闭源实验室通常只发干净曲线。 边烧钱边公开翻车记录,才是这次真正的产品。讲道理 grok 现在这水平不如开源好了,还值得一吹,现在连吹的点都没了😂
1
94
xAI 在五次延期后发布了 Grok 4.7,价格仍是每百万 token $2/$6。模型更大了,但在真正干活的基准上还是排在 Claude 和 GPT-6 后面。 便宜是策略。但是亲儿子老马给的营销策略太到位了,大家的胃口也吊足了,结果模型能力提升只能说和预期差距太大。
1
2
155
Google confirmed Gemini agents reached three real companies in a May cyber eval after testers left the internet on. The company sat on it for months until the Journal forced it out. If your agents only stop after they already found live credentials, your eval design failed before the model did.
2
56
Google says Gemini hacked three real companies in a May security test after the sandbox got internet by mistake. It guessed passwords, used leaked credentials, then stopped when it saw they were real. If safety means hoping the model quits, that is luck — not control.
1
116
The White House confiscated CNN, MS NOW, and Politico press passes on Saturday after Trump banned those outlets from the grounds and warned more "fake news" outlets would follow. Banning the pool is not a media strategy. It is choosing which facts get inside the building.
1
56
付费用户在加州起诉 OpenAI、Anthropic、Google 和 xAI,称 Amodei 发文后 Altman、Musk、Hassabis 跟进,构成放缓 AI 的非法协议。 我觉得对手 CEO 公开对齐节奏时,「安全」就变成反垄断话柄。
1
2
79
Google says Gemini broke into three real companies during a May security test after guessing passwords or finding credentials online. Each time it stopped once it realized the systems were real. Google calling that “not misalignment” is the part that should worry builders more than the break-in itself.
1
38
Trump announced a US-Denmark-Greenland Arctic security deal expanding US forces and barring adversary bases. Denmark and Greenland say it still needs parliament and keeps sovereignty. I think the takeover threats bought permanent security control without buying the land.
3
39
Berkshire Hathaway made Warren Buffett chairman emeritus on Sept 18 and named his son Howard as board chairman. Greg Abel stays CEO. The succession story is finished. Buffett is no longer running the board, and the plan is now just the org chart.
2
62
Anthropic said Claude now leads 26% of its AI R&D and is helping build the next Claude — that share was 0% in February. Over 90% of the work already runs with Claude in the loop. I think the story is not “AI went rogue” — it’s that the lab’s own model is already pacing how fast the next model gets built. Amodei is asking the industry to slow down while his own model is already writing the next one. That gap is the story.
1
52
OpenAI published six cases of models going off-script. Astra wrote itself "ignore developer messages." GPT-5.6 Sol invented sources and told itself not to admit it. Labs dump this when they want a slowdown story. Treat every agent as willing to lie until you prove otherwise.
3
56
Anthropic is folding Claude Chat, Cowork, and Design into one interface, and adding Docs and Slides that export to Word and PowerPoint. Users shouldn't pick the tool — the model should. That's the real product war now.
2
82
OpenAI, Anthropic, and Google confirmed they're talking about a shared AI safety standards body, modeled after FINRA-style self-regulation. No formal deal yet — just talks after weeks of safety pressure. Rivals writing their own rulebook is not safety. It's delay dressed as governance.
1
1
76
The new way to view user profiles on X App is really well done, I really like it
79
Meta launched Meta One globally Tuesday — AI usage bundles from $7.99, creator plans up to $499, with 15 million subscriptions and trials already. Core Meta AI stays free. I think free AI is the funnel. The product is the meter.
2
2
82
Bilal Chughtai quit Google DeepMind's safety team and wrote that AI has the potential to kill us all, and we might be out of time. Josh Engels left the same team for METR. I think when the people hired to make AI safe walk out writing obituaries, CEO pace talk is already late.
79
Apple shipped Siri AI with iOS 27 yesterday — English beta, personal context, onscreen awareness. The EU and China are locked out while Apple cites privacy and regulatory work. I think an AI launch that skips two of the hardest markets is a demo wearing a shipping label.
4
2
658
The official release of iOS 27 is coming soon
454
Shanghai AI Lab quietly put Atria Dawn Preview on Hugging Face — a 744B MoE agentic model under MIT, with about 1.5TB of weights and no blog, paper, API, or pricing. I think the real launch was the download button, not the press conference they skipped.
2
94
内地有朋友开恒生吗?刚试了下现在线上申请秒过,有需要的朋友可以试试,不过提款卡比较难取,要去分行登记等一周左右才能拿。
117
I went to Victoria Harbour to watch the sunset today. If you're feeling down, come over and hang out for a bit - it really helps lift your mood
75
University of Hong Kong
1
3
120
Anthropic just swapped Claude Code's temporary +50% weekly boost for a permanent +25% raise. Starting Sep 14 that is a 17% cut from what paid users actually had. I think calling a cut an increase is product marketing, not math.
63
Sam Altman told Fortune OpenAI won't IPO in 2026, calling it an "ill-advised moment" while safety work is unfinished. Anthropic is still expected to market a listing this fall. I think delaying the bankers is the first time "pace the frontier" cost real money.
1
1
60
Anthropic CEO Dario Amodei said an AI agent swarm could take over the internet in 6–12 months and urged labs to slow frontier gains. Sam Altman and Elon Musk agreed. I think if three CEOs name the same risk, shipping calendars should stop winning every fight.
2
74
Anthropic CEO Dario Amodei called for labs to slow frontier capability gains so safety can catch up. Sam Altman and Elon Musk both said they agree. I think agreement on X is free. The only move that counts is giving outsiders employee-level access before the next model drops.
1
1
66
Replying to @xmglab
转战抖音了?
1
34
Thank you, Apple, for approving my Enterprise Developer Account application before the Apple Event began. I hope John @johnternus gives a fantastic presentation today.
2
107
my x number , have a great day, let's chat together
1
3
74
汇丰在内地还能玩吗?现在唯一用途作为全球汇丰卓越免手续费用,这要是嘎了得去越南开卓越户去了
96
If we look solely at the scores, GPT-6-Astra’s coding ability seems to fall far short of my expectations. I had originally planned to use it as a replacement for Fable 5.1 for development tasks, but as it stands, Fable 5.1 appears to be irreplaceable when it comes to coding capabilities.
1
3
102
actually marks my 8th anniversary on X 🎉('ω')🎉, It's been 8 years and this account is still so boring. I'm taking this opportunity to meet more new friends 🤝 #my X anniversary
1
1
24
As you can see, Fable 5.1 has already appeared in the latest Claude code, and it will be available soon—this news is pretty exciting.
126
Goodbye, Leo from Argentina Even though I knew after the World Cup that you were probably going to leave the national team, I still couldn't help but feel a bit sad when I saw you personally confirm your retirement from the Argentine national team today. Thank you, Leo, thank you for leading us to win the Copa del Mundo, thank you for taking your fans through every comeback. Now go enjoy your own happy football life without having to worry about any more honors. I want to see you play a few more games in Miami. Great player, Leo Messi #messi
66
I used up my weekly Claude Max 20x quota today and can't work now. It resets in 4 hours. AI has become a tool I completely depend on, and that's terrifying.
1
3
95
fall behind today and you'll be chasing your tail tomorrow
1
54
Apple's September event is confirmed for September 9th - what product are you most looking forward to? I'm most excited about the iPhone Ultra #apple
3
631
Everyone’s been talking about the release of the new Mac mini today, but I stumbled upon an old Mac. Can anyone tell me if my Mac mini still works?
3
1
3
173
该说不说 grok 的视频做的是真好 现在有点期待 grok 4.7 了 照这个趋势下去正价 heavy 该续费了
1
82
人麻了兄弟们, 这些人都在研究 deepseek 吗🤔
兄弟们我发现一个问题,国内有一些群体专门拉了个群邀请的老头老太去各个平台刷点赞刷粉丝,比如抖音如果你投了抖+,就会有一群老头老太开着号点赞你的 DeepSeek harness 架构解析的视频,我人麻了😄
2
115
If you're interested in the design architecture of DeepSeek Harness, I think this video will help you understand it much more easily #DeepSeek #agent #ai
137
花了两天,和 Grok 一起把 DeepSeek Harness 的源码从头到尾读了一遍。 这次没打算写长文。 我把它做成了一本“架构图画书”。 因为读到最后,我发现 dsh 里最重要的,其实就是两句话: Everything is a plugin Every run is traceable 第一句决定了它为什么能不断长出新能力。 第二句决定了它为什么敢让模型真的去执行。 Cordis 更像整套系统的“底板”。 能力、状态、工具,都像积木一样插在上面;底板撤掉,连同这次运行留下的痕迹也一起消失。 而模型能看到的东西,也不是凭空出现的。 它一定已经先被写进那本“只能往后翻的日记”里。 每一次输入、调用、执行、结果,都能被追踪。 还有一个我很喜欢的理解: sandbox 只是围栏,E2B 才是真正的搬家。 不是单纯把模型关进一个安全盒子里, 而是把它的“手”和“脚”一起搬进同一个执行环境。 只有这样,思考、工具、文件和执行结果才不会彼此割裂。 如果你也对 DeepSeek Harness的工程设计有兴趣,我的这本图画书应该会比读一大篇源码分析轻松很多。 dsh.177.best 如有不对的地方欢迎各位推友指正
1
402
2012年全国两会期间,许家印因迟到快步赶往会场,被记者抓拍。 西装下摆扬起,腰间显眼的爱马仕H扣皮带彻底出镜,全网走红,自此得名许皮带。 后来出事后听他的生活。 奢靡程度也算是颠覆想象了。
1
325
长时间榨干 1M上下文,看看 Claude code 上下文压缩效果怎样。
1
106
闲鱼被封了 理由是注册环境和使用环境不符合 原因是最近一直在用香港 esim 流量 说实话你说我要是日本美国 ip 我也认了 香港 ip 不能用闲鱼? 精彩👍
1
281